Ilya Kurdyukov [Sat, 13 Nov 2021 11:22:14 +0000 (18:22 +0700)]
faster vp8_regular_quantize_b_sse4_1
Gives 10% faster VP8 encoding in simple tests.
This patch requires testing on wider datasets and encoder
settings to see if this speedup is achieved on most data.
Johann [Tue, 16 Nov 2021 21:02:15 +0000 (06:02 +0900)]
MacOS 12 is darwin21
Remove -mmacosx-version-min. The library does not use
any calls which are affected by the platform version.
There is also no version 10.16 as it went from 10.15
to 11 and now to 12.
At some point it may be good to clarify that the bare
-darwin- target is for iOS and the -darwinN- targets
are for macOS.
Mikko Koivisto [Mon, 15 Nov 2021 18:47:05 +0000 (18:47 +0000)]
vp9: Fix multiplication overflow
Fix UBSan error reported from aosp Cuttlefish device:
/vp9/encoder/vp9_ratectrl.c:238:33: unsigned integer overflow: 2500000 * 1800 cannot be represented in type 'unsigned int'
...by casting the operand and the result of multiplication
to 64bit integer.
Test: vp9 webrtc streaming with Cuttlefish
Change-Id: Id5bb3d4071a96179caffae0829d3cc4e48c7614b
James Zern [Sat, 6 Nov 2021 23:48:13 +0000 (16:48 -0700)]
mem_sse2.h: loadu_uint32 -> loadu_int32
this changes the return to int32_t which matches the type with usage
of this call as input to _mm_cvtsi32_si128(), _mm_set_epi32(), etc.
fixes implicit conversion warning with clang-11 -fsanitize=undefined
James Zern [Sat, 6 Nov 2021 23:43:11 +0000 (16:43 -0700)]
mem_sse2.h: storeu_uint32 -> storeu_int32
this changes the parameter to int32_t which matches the type with usage
of this call using _mm_cvtsi128_si32() as a parameter. quiets an
implicit conversion warning with clang-11 -fsanitize=undefined
test/video_source.h:194:33: runtime error: implicit conversion from type
'int' of value -32 (32-bit, signed) to type 'unsigned int' changed the
value to 4294967264 (32-bit, unsigned)
James Zern [Wed, 3 Nov 2021 00:19:10 +0000 (17:19 -0700)]
tools_common.h: add VPX_TOOLS_FORMAT_PRINTF
and use it to set the format attribute for printf like functions. this
allows the examples to be built with -Wformat-nonliteral without
producing warnings.
James Zern [Tue, 2 Nov 2021 23:29:52 +0000 (16:29 -0700)]
vpx_codec_internal.h: add LIBVPX_FORMAT_PRINTF
and use it to set the format attribute for the printf like function
vpx_internal_error(). this allows the main library to be built with
-Wformat-nonliteral without producing warnings; the examples will be
handled in a followup.
James Zern [Fri, 1 Oct 2021 20:46:02 +0000 (13:46 -0700)]
{vp8,vp9}_set_roi_map: fix validation with INT_MIN
previously ranges were checked with abs() whose behavior is undefined
with INT_MIN. this fixes a crash when the original value is returned and
it later used as and offset into a table.
James Zern [Fri, 1 Oct 2021 20:46:02 +0000 (13:46 -0700)]
{vp8,vp9}_set_roi_map: fix validation with INT_MIN
previously ranges were checked with abs() whose behavior is undefined
with INT_MIN. this fixes a crash when the original value is returned and
it later used as and offset into a table.
Define VPX_NO_RETURN as __declspec(noreturn) for MSVC. See
https://docs.microsoft.com/en-us/cpp/cpp/noreturn?view=msvc-160
This requires moving VPX_NO_RETURN before function declarations because
__declspec(noreturn) must be placed there. Fortunately GCC's
__attribute__((noreturn)) can be placed either before or after function
declarations.
Define VPX_NO_RETURN as __declspec(noreturn) for MSVC. See
https://docs.microsoft.com/en-us/cpp/cpp/noreturn?view=msvc-160
This requires moving VPX_NO_RETURN before function declarations because
__declspec(noreturn) must be placed there. Fortunately GCC's
__attribute__((noreturn)) can be placed either before or after function
declarations.
Yunqing Wang [Fri, 23 Jul 2021 17:55:10 +0000 (10:55 -0700)]
Disable allow_partition_search_skip feature
This feature was added to help speed up still images and slideshows.
It didn't work anymore, and thus was disabled. Code cleanup will
follow.
This had negligible impact to regular test sets. Borg test result
on ugc360p set at speed 3.
avg_psnr: ovr_psnr: ssim: speed:
-0.244 -0.278 -0.153 -0.973
target_bandwidth is int64_t, but layer_target_bitrate[0] is an int. this
is safe in the only place it's set because target_bandwidth defaults to
1000. target_bandwidth is later used to populate the cpi's target, which
is an unsigned int so there may be further fixes/cleanups that can be
done.
Jorge E. Moreira [Wed, 30 Jun 2021 18:33:51 +0000 (11:33 -0700)]
Avoid overflow in calc_iframe_target_size
The changed product was observed to attempt to multiply 1800 by 2500000,
which overflows unsigned 32 bits. Converting to unsigned 64 bits first
and testing whether the final result fits in 32 bits solves the problem.
Cheng Chen [Thu, 17 Jun 2021 22:36:18 +0000 (15:36 -0700)]
Disallow skipping transform and quantization
The encoder has a feature to skip transform and quantization based
on model rd analysis. It could happen that the model
based analysis lets the encoder skips transform and quantization, while
a bad prediction occurs, leading to bad reconstructed blocks, which
are intrusive and apparently coding errors.
We add a speed feature to guard the skipping feature.
Due to the risk of bad perceptual quality, we disallow such skipping
by default.
On hdres test set, speed 2, the coding performance difference is 0.025%,
speed difference is 1.2%, which can be considered non significant.
Marco Paniconi [Fri, 25 Jun 2021 06:34:36 +0000 (23:34 -0700)]
vp9-rtc: Extract content dependency in cyclic refresh
For usage in the external RC. When content_mode = 0,
the cyclic refresh has no dependency on the content
(motion, spatial variance, motion vectors, etc,).
The content_mode = 0, when compared to content_mode = 1,
on rtc set for speed 7: has some regression on some
clips (~3-5%), but overall/average bdrate loss is
about ~1-2%.
Comparing aq_mode=3 with content_mode = 0, vs aq_mode=3:
about ~14% avg/overall bdrate gain, but has ~3-7% regression
on some hard motion clip (e.g.m street).
projects / libvpx / log